This is a preliminary evaluation of forecasts made for the German Forecast Hub. These evaluations are preliminary - this means I cannot currently rule out any mistakes and the plots and analyses are subject to change. The evaluations are not authorised by the German Forecast Hub team.
Feel free to reproduce this analysis. To that end, you can clone this repository and the Forecast Hub repository, specify the directory of the submissions files in the R-Markdown script and run the script.
If you have questions or want to give feedback, please create an issue in this repository.
Overview of all the models and metrics
Weighted interval score broken down into its components “overprediction penalty”, “underprediction penalty” and “sharpness”
|
| | 0% |
|…………………………….. | 50% ordinary text without R code
|
|…………………………………………………………….| 100% label: unnamed-chunk-3 (with options) List of 1 $ results: chr “asis”
Plot of the percentage of true values captured by each prediction interval
Plot of the percentage of true values below each predictive quantile level